Everything is connected: Inference and attractors in delusions.

Area of Science:

• Computational Psychiatry
• Neuroscience
• Artificial Intelligence

Background:

• Delusions, characterized by certain false beliefs resistant to evidence, appear contrary to Bayesian inference principles.
• Schizophrenia is associated with reduced certainty in the brain's world model, posing a paradox for Bayesian brain theories.

Purpose of the Study:

• To model the emergence of delusions using an active inference framework.
• To investigate how changes in model parameters influence belief certainty and delusion formation.
• To explore the potential of computational models to simulate therapeutic interventions.

Main Methods:

• Utilized an active inference Markov decision process model, simulating a Bayes-optimal decision-making agent.
• Manipulated model parameters, specifically confidence in sensory input and action-based states.
• Incorporated affect into the model to examine its influence on social inferences.

Main Results:

• Moderate parameter changes, decreasing sensory confidence and increasing action confidence, induced delusions.
• Inclusion of affect amplified delusions, particularly in social contexts.
• The model replicated psychological phenomena like choice-induced preference change and optimism bias.
• Delusions emerged from conditional dependencies creating 'basins of attraction' for beliefs, not single parameter changes.
• Simulated antidopaminergic antipsychotics (reducing action confidence) enabled the model to escape these attractors.

Conclusions:

• Computational modeling provides a framework for understanding delusion formation in terms of aberrant Bayesian inference.
• Altered confidence in sensory evidence versus self-generated actions can mechanistically explain delusions.
• The model offers a testable hypothesis for how antipsychotics may alleviate delusions by altering action confidence.